1. Lu Xing
CS59000GDM
9/21/2018

2. Challenges for RDF data management
Scalability:
Cause of problem: RDF data are stored in triples and manages by RDBMS.
Single-machine Systems
Distributed systems
SW-Store
SHARD
Hexastore
YARS2
RDF-3x
Virtuoso
Expensive join operations and large intermediate results!

3. Challenges for RDF data management
Generality:
Cause of problem: general purpose queries on RDF data is not supported

4. Highlights of Trinity.RDF
Distributed system
Native graph
In-memory graph
Smaller intermediate results

5. Join vs. Graph Exploration -- Join
RDF data is scanned to generate bindings

6. Intermediate results of Join operation
?director
?award
?movie
?actor
?movie_award
J_Cameron
Oscar_Award
Titanic
L_Dicaprio
Best_Picture
Avatar
S_Worthington
G_Lucas
Saturn_Award
P_Haggis
Crash
D_Cheadle
Star War VI
M_Hamill

7. Join vs. Graph Exploration -- Graph

8. Join vs. Graph Exploration -- Graph

9. Join vs. Graph Exploration -- Graph

10. Join vs. Graph Exploration -- Graph
No need to explore nodes here.

11. Join vs. Graph Exploration -- Graph
Why do the queries execute in this way?

12. Trinity: A Distributed Graph Engine on a Memory Cloud (Trinity supports fast graph exploration as well as efficient parallel computing)

13. Typo

14. How to model graphs?

15. How to model graphs?
Problem: What if node n has a huge number of neighbors, but none of them are in the same machine as n?

16. Graph partition
The in-adjacency-list stored in machine i

17. Other auxiliary data structures and operators
Indexing predicates
Local predicate indexing
Global predicate indexing
Basic graph operators

18. Query process Decompose Q into an ordered sequence of triple patterns
Find matches for each qi
From each match, explore the graph to find matches for qi+1
Gather the matchings for all individual triple patterns to the centralized query proxy

19. Single triple pattern matching
From subject to object

20. Single triple pattern matching
All the possibilities of src

21. Single triple pattern matching
Get the key nid to obtain node’s
neighbors in machine i

22. Single triple pattern matching
Select list of nodes who have p as predicate

23. Single triple pattern matching

24. Example for Single Pattern Matching

25. Example for Single Pattern Matching

26. Example for Single Pattern Matching

27. Multiple Pattern Matching
No joins! Eagerly prune invalid matchings.

28. The order of queries matters
Query Q: q1, q2, q3, …, qn
RDF data … q1 q2 qn

29. Exploration Plan Optimization

30. Define Exploration Point
A node that does not contain any redundant values. Hence it introduces fewer redundant values for other variables in the exploration.
Exploration points
This is how a graph grows.

31. Use Dynamic Programming to compute
Start from subgraph of size 1
Fill the 2-D table with the minimum cost

32. Evaluation – Running time

33. Evaluation – Running time (cont’)

34. Thank you!

35. Single triple pattern matching
Otherwise, src is a constant

36. How about cost?
The cost is a linear combination of the size of the results and bindings of src.